Research On Pathological Image Recognition Technology Based On Deep Learning

Deep learning is one of the fastest developing technologies in recent years.All kinds o f artificial intelligence technologies based on deep learning have been integrated into every corner of life.The application of deep learning in the field of medical image and the realiz ation of computer-aided diagnosis can not only improve the diagnosis efficiency of doctors,but also ensure the accuracy of their diagnosis.Based on the analysis of traditional image processing technology and deep learning theory,combined with the characteristics of patholo gical images,this paper studies the recognition and segmentation technology of gastric canc er pathological images.The main contents of this paper are as follows:(1)The gastric cancer pathological image recognition based on migration learning is rea lized.We analyzed the advantages and disadvantages of the related technologies such as dee p learning and image processing,and proposes to use the transfer learning to complete the pre training and initialization of the network.By contrasting and analyzing the results with the same model without pre training,experiments show that the recognition algorithm based on the transfer learning is much more accurate than the non migrating algorithm.The acc uracy of Res Net-50 network based on transfer learning achieved 98.7% and 95.1% in traini ng set and test set.(2)We proposed two segmentation algorithms for cancer cells: one based on CNNs(Con volutional Neural Networks)and the other based on U-Net.Based on the research and sum mary of common image recognition technology,the algorithm based on CNN proposes to di vide the image into several grids,classify each grid separately,and finally summarize all th e classification results to segment the image.The final performance of the algorithm is poo r,so we proposed to use FCNs(Full Convolution Neural Networks)to segment cancer tissue area,but considering the poor performance of ordinary FCNs to achieve pixel-level segme ntation,we finally choose an improved fcns model U-Net to achieve segmentation task.The algorithm based on U-Net network has better segmentation performance,the segmentation accuracy has reached 79.91%.